sign in sign up
<<
Home , Nexus S

How is Energy Consumed in Display Applications? Xiang Chen, Yiran Chen Zhan Ma, Felix . A. Fernandes ECE Department, University of Pittsburgh Telecommunication America Pittsburgh, PA, USA 15261 Richardson, TX, USA 75082 +1 (412) 624-5836 +1(972)761-7134 {xic33, yic52}@pitt.edu {zhan.ma, ffernandes}@sta.samsung.com

ABSTRACT of these applications depends heavily on the type of display have emerged as a popular and frequently used technology being used, which also plays an important role in platform for the consumption of multimedia. New display human-machine interaction. Many researches have already studied technologies, such as AMOLED, have been recently introduced to the power and performance of different display technologies [1]. smartphones to fulfill the requirements of these multimedia Very recently, AMOLED (Active-Matrix Organic Light Emitting applications. However, as an AMOLED screen’s power Diode) display panels have begun to replace conventional LCD consumption is determined by the display content, such (Liquid Crystal Display) technology in mainstream smartphones. applications are often limited by the battery life of the device they Compared to LCD, AMOLED offers much better display quality are running on, inspiring many researches to develop new power and higher power efficiency because of its unique lighting management schemes. In this work, we evaluate the power mechanism. This unique property has led to much research consumption of several applications on a series of Samsung involving the power evaluation and modeling of AMOLED as well smartphones and take a deep look into AMOLED's power as the performance variance among different AMOLED panel consumption and its relative contributions for multimedia apps. We designs. However, due to the large variety of AMOLED panel improve AMOLED power analysis by considering the dynamic designs and the fast pace of smartphone development, factors in displaying, and analyze the individual factors affecting most of this research only aims at a particular smartphone device power consumption when streaming video, playing a video game, or application. It is not clear whether there is any significant power and recording video via a device’s built-in camera. Our detailed efficiency improvement between different generations of measurements refine the power analysis of smartphones and reveal AMOLED display technology or if it is possible to obtain an some interesting perspectives regarding the power consumption of overview display power efficiency under different applications. AMOLED displays in multimedia applications. In this work, we evaluate the power consumption of the displays of Categories and Subject Descriptors several high-end Samsung smartphone models. In addition to this, we discuss the power modeling of different generations of K.6.2 [Management OF Computing and Information Systems]: AMOLED screens and conduct a detailed power analysis of several Installation Management – Performance and usage measurement. popular multimedia applications. Based on the data that was General Terms collected, we draw some interesting conclusions which may Algorithms, Management, Measurement, Performance, Design. influence future research in the field. For example: • Subsequent generations of AMOLED products do not yield the Keywords significant per-unit improvement in power efficiency that was OLED display, Smartphone, AMOLED, Video power. expected. The power difference between AMOLED products is 1. INTRODUCTION mainly realized by design metrics like size and sub-. Smartphones now play a large role in almost every aspect of our •The power consumption of chromatic color can be efficiently daily lives, being utilized in activities such as communication, reduced by implementing a dynamic color tuning technique. personal planning, and entertainment. Although the complexity and • The power consumed during the video decoding process is capabilities of these devices continues to grow at an amazing pace, responsible for only a small portion of the total power required for smartphones are now expected to continually become lighter and the end user to actually view the video. slimmer. When combined with power-hungry multimedia applications, the limited battery capacity allowed by these •In video games, the AMOLED display’s power consumption expectations now motivates significant investment into smartphone varies greatly, and it is relatively small in some cases when power management research. compared to the overall system power consumption. Multimedia applications, such as streaming video players, games, •The power consumed by the AMOLED during video recording is and image and video capturing tools, now comprise a considerable relatively small compared to the overall power consumption. portion of the daily usage of smartphones. The power consumption The remainder of our paper is organized as follows: Section 2 Permission to make digital or hard copies of all or part of this work for presents previous related work; Section 3 gives our adopted personal or classroom use is granted without fee provided that copies are methodology for power evaluation, including the devices under test not made or distributed for profit or commercial advantage and that copies (DUTs) and the test environment setup; Section 4 models the power bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific consumptions of several different AMOLED panels on Samsung permission and/or a fee. smartphones; Section 5 presents the power analysis’ of video ACM HotMobile'13, February 26-27, 2013, Jekyll Island, Georgia, USA. streaming players, video games, and camera recording, Copyright 2013 ACM 978-1-4503-1421-3 ...$10.00. respectively; Section 6 concludes our work. 2. RELATED WORK Panel Smartphone Resolution Since its development, AMOLED technology has been the topic of Size a great deal of research, much of which has attempted to describe 4.0” 480*800 and model its power characteristics. The authors of [2] presented a Galaxy S2 4.5” 480*800 classic AMOLED power modeling in smartphone, which is Galaxy 4.65” 720*1280 adopted in this paper. The authors of [3] evaluated the AMOLED Nexus power model’s accuracy in practical performance and helped Galaxy S3 4.8” 720*1280 application developers to improve the energy efficiency of their smartphone applications. OLED power model is part of the system (a) (b) power model. The display modeling of AMOLED has also been Fig. 1. (a) AMOLED sub-pixel matrix; (b) Smartphone integrated into a system level power monitor and analyzer, as done AMOLED screen specs. in [8] and [9]. contrast experiments. Power optimization schemes based on AMOLED displays have We adopted a mobile device power monitor produced by Monsoon also been studied from the applications point of view. The authors Inc. for real time power measurement. It supports a sampling of [15] reduced the power consumed while displaying an arbitrary frequency up to 5 kHz and can be used to replace the battery as the image through the use of OLED’s dynamic voltage scaling (DVS) power supply to the smartphone. The power monitor is directly technology, while the authors of [4] extended this DVS scheme into connected to the smartphone and records a power histogram. video streams. The authors of [7] extended the dynamic gamma During the power evaluation of the display modules, we disabled correction power saving scheme for video games to the AMOLED all unnecessary system services that may have caused any platform. The author of [5] shows OLED screens contribute significant power consumption noise, including and Wi-Fi significantly to the energy consumption of web browser and then network communication, background services, and power applies the optimization techniques from [6] to reduce it. These optimization applications. The contrast experiments are designed to works pointed out that, while a smartphone spends most time idle separate the power consumption of display from that of the system, when web browsing, streaming video, playing video game, and e.g., CPU and GPU. Most power consumption test are completed camera recording, the smartphone demands significant system while the device’s screen brightness is set to maximum. The resources in a continuous manner. However, we find the power specific test bench details will be presented in later sections. contribution made by the OLED screen can be small. 4. AMOLED DISPLAY EXPLORATION 3. METHODOLOGY 3.1 Devices Under Test (DUTs) 4.1 AMOLED Power Evaluation on Sub-pixel Different from previous research, instead of directly measuring and We examined five Samsung products, including the Nexus S then comparing the power consumption of each entire screen, we (released in 2010), the Galaxy S1, Nexus, and S2 (released in attempt to normalize the displays and compare the per-unit power 2011), and the Galaxy S3 (released in 2012). During the tests, the efficiency for different AMOLED products. Because of this we devices ran the Android and a suite of test found that, in contrast to the conclusions drawn by previous applications. The screens of all the tested devices are all based on research, when comparing the power efficiency between different AMOLED technology, though they are built with different display generations of AMOLED technology: sizes, resolutions, and technology generations, i.e., Super AMOLED, Super AMOLED Plus and Super AMOLED HD, Subsequent generations of AMOLED products do not yield the respectively. These screens well represent the evolution of significant per-unit improvement in power efficiency that was Samsung's OLED technology in recent years. expected. The power difference between AMOLED products is mainly realized by design metrics like size and sub-pixel matrix. 3.2 Power Evaluation Setup Many smartphone power models have been proposed in recent A smartphone display is composed of many individual . The research and are included in the embedded Android power monitor color of a pixel is constructed via the combination of a few basic [8][9][10]. However, these models generally have limited colors (e.g., red, green, blue, or RGB), which themselves are the adaptability to the many variations of available hardware only colors directly emitted from the display unit via tiny sub- configurations, incurring inevitable run-time evaluation errors [11]. pixels. The display unit is defined by the arrangement of these sub- Rather than calculating power information from software, we pixels, which is known as the sub-pixel matrix design. As shown in utilized an external power monitor to directly measure the battery Fig. 1(a), in the AMOLED displays we tested, every pixel has three behavior for power consumption breakdown with a series of sub-pixels corresponding to the basic colors of RGB color space. Unlike LCD technology, the power consumption of an AMOLED

600 600 600 600 R G B R G B R G B R G B 500 500 500 500 R' G' B' R' G' B' R' G' B' R' G' B' 400 400 400 400

300 300 300 300

200 200 200 200

100 100 100 100

Power Consumption Consumption Power (mW) Power Consumption Consumption (mW) Power 0 Consumption Power (mW) Power Consumption Consumption (mW) Power 0 0 0 0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250 Grey Level (0~255) Grey Level (0~255) Grey Level (0~255) Grey Level (0~255) (a) (b) (c) (d) Fig. 2. AMOLED screen power consumption: (a) Nexus S; (b) Galaxy S2; (c) ; (d) Galaxy S3 2 Solid lines are the original overall screen power consumption, dash lines are the screen power consumption normalized to 4 inch . screen is highly color dependent. The power consumption of each 1200 Practical White - Full Brightness pixel varies with the chromatic color composition that is being 1000 displayed, as not only do more intense colors require more power Practical White - Half Brightness 800 to display, but each individual sub-pixel has its own unique power Theoretical White - Full Brightness distribution curve. Hence, various sub-pixel matrix designs are 600 proposed to balance the lighting efficiency and display quality by Theoretical White - Half Brightness 400 adopting different alignments, area-to-area ratios, and etc. of individual sub-pixels. Fig. 1(a) illustrates the contrasting sub-pixel 200 matrix designs of Super AMOELD Plus (based on identical RGB Power Consumption Consumption (mW) Power 0 strips) and Super AMOLED (based on PenTile design) [12]. 0 50 100 150 200 250 Grey Level (0~255) In many prior works involving AMOLED screens, the pixel-level power modeling is often expressed as: Ppixel =f(R)+h(G)+g(B). This Fig. 3. System chromatic color tone remapping. models the power consumption of a pixel as the summation of the to different power models. power consumption of the individual RGB sub-pixels [2]. f(), h(), and g() are the color component dependent functions and R, G, and 4.2 Dynamic Color Tuning B are the input signals, which are usually represented as a grey level Based on the color power measurement, we also found that: of [0, 255]. In Samsung OLED products, a standard gamma correction of 2.2 is applied to the input signals in order to improve In the operation of an AMOLED screen on a smartphone, the actual the display quality. Thus, the relationship between the power displayed color and intensity may not be representative of the consumption of the AMOLED screen and the grey level becomes original raw RGB composition being sent to the screen as it may nonlinear, as shown in Fig. 2. have been modified by a dynamic color tuning system in order to minimize power consumption. Fig. 2 also shows that the power models of different display modules vary significantly. One example of this is the difference The previous single color evaluation is realized by adjusting the between the Super AMOLED or PenTile AMOLED, used in grey level of individual RGB color component while disabling the Galaxy S3 and the Super AMOLED Plus used in Galaxy S2: the other display channels. However, when we measured the chromatic power consumption of the PenTile design in Super AMOLED is color’s power consumption in Galaxy S3, we found that it is not much more balanced among the different colors than the traditional simply the summation of the individual RGB channels’ pixel power RGB strips utilized in Super AMOLED Plus. In addition to this, model, which was discussed in Section 3.1. display panels with the PenTile design also show an averagely We examined the power consumption of the AMOLED screen lower power efficiency than traditional RGB strip. However, as when its color changes from black to white. At every measured PenTile design aligns more low power consumping green sub- color composition, the same grey levels are applied to each sub- pixels in the pixel matrix, it’s supposed to be more power efficient. pixel. As shown in Fig. 3, at a grey level of 100 and above, the Hence, we have to also take the display panel size into theoretical power consumption that is calculated by the existing consideration to compare the power efficiency. simple power model is 7% to 14% higher than the practical In many existing AMOLED power measurements and models, the (measured) power consumption under full brightness and 9% to power consumption of a display panel is evaluated as a whole. The 22% higher under half brightness. Similar phenomena were many different panel sizes and sub-pixel alignments available today observed when displaying arbitrary chromatic colors, such as aqua, make it very difficult to conduct a fair comparison of the power pink, purple, and etc. efficiency between different AMOLED technology generations. In practice, it is very difficult to measure the power consumption of Therefore, we focused on analyzing the sub-pixel power efficiency. an AMOLED screen in the 24-bit RGB color space and derive a To accomplish this task, we first normalized the area of each generic power model. However, it is still possible to integrate a individual sub-pixel to account for the size difference between sub- simplified model into the smartphone for AMOLED screen power pixels in PenTile technology. The power consumption of the entire optimization. In fact, in Samsung’s display system, an image AMOLED screen while displaying each basic color at every grey processing engine called MDNIe is implemented by a designated level was then recorded and scaled to represent a screen size of 4 chipset just in front of the graphic buffer to the display panel [13]. inch2. The normalized power consumption of the sub-pixels in each One application example of MDNIe is dynamic color tuning: the scaled display is depicted as the dashed lines in Fig. 2. color composition being displayed on the screen can be adjusted by MDNIe to decrease power consumption while maintaining contrast Although some prior work has claimed that the power efficiency of levels. Hence, this system integrated dynamic color tuning has AMOLED screens has as much as doubled between generations, achieved power savings by color tuning effectively. i.e., from the Galaxy S2 to the Galaxy S3 [18], we found that the practical power improvement is not nearly that significant. For 5. POWER ANALYSIS ON DISPLAY example, when comparing to the oldest AMOLED screen (in the RELATED SMARTPHONE APPLICATION Nexus S) to the latest (in the Galaxy S3), a jump of two generations, the highest pixel-level power efficiency is between only 5.7% and 5.1 Video Power Performance 29.5% improved at a grey level of 50. Interestingly, we found that Although OLED technology has substantially improved the power the most significant factor that contributes to the power efficiency of displays when compared to traditional LCD screens consumption difference between the different smartphone models [14], the display panel is still one of the most power-consuming is the sub-pixel area ratio. For example, although all the Super components in a smartphone [15]. Applications with dynamic AMOLED display panels follow G-R-G-B sub-pixel matrix design, display contents, e.g., streaming video player, are considered as the sub-pixel area ratio is slightly adjusted in the different models being energy-hungry by previous research. to accommodate the specific screen size and , leading However, in this work, we found that: In video stream player, the AMOLED screen’s power consumption 0.6 Mbps 1.2 Mbps 1.8 Mbps is highly content dependent. The video decoding process 1600 contributes very marginal power consumption to the application. 2.4 Mbps 3.0 Mbps 1200 We examined a set of pure video streams without audio track and analyzed the composition of the application’s power consumption. 800 We also conducted a breakdown of the screen’s power consumption over each functional component. To better represent 400

typical display content differences, we followed YouTube’s video Consumption Power (mW) category and selected four types of video streams: music videos, 0 sports, gameplay videos, and news reports. All of the video clips 0 200 400 600 800 Time (s) tested are selected and downloaded from YouTube. Most of these Fig. 4 System chromatic color tone remapping video streams are between 2-3 minutes in length and vary between 1000 and 2500 frames. tuning. As a comparison, the least power consuming video content category is music video, which consumes only 15% and 17% of 5.1.1 Power consumption with different display contents total smartphone power at 60fps and 30fps, respectively. Because of the previously discussed color-dependent power model associated with AMOLED pixels, the display content directly Our results also show that the power consumption of the AMOLED determines the power consumption of an AMOLED screen. We screen decreases considerably when the resolution of the displayed evaluated 50 local video streams in each category on a Samsung video stream degrades: the ratio of the power consumed by the Galaxy S3 in the field study. All of the video streams were encoded AMOLED screen to that of the entire smartphone decreases to with a bit rate of 0.8 Mbps without a sound track. The refresh frame between 7% and 15%. This indicates that while streaming an online rates and display areas were configured differently to evaluate the video, AMOLED screens consume quite a small portion of the system decoding performance. The overall power consumption of overall smartphone power, especially when compared to the the device was recorded while at the same time monitoring the potential power cost of network, which can require up to 2W. individual component power breakdown. Moreover, we simulated approximately 200 video streams under The average power consumptions of each category of video are each category to obtain the power consumption statistics of shown in Table 1. In full screen tests, videos are stretched to fill the AMOLED screens. Our simulated results show that power entire screen, while in original size tests the videos are displayed consumption of the AMOLED screen is between 74.6mW and on the screen in a letterbox with a resolution of 640x360 (the rest 374.2mW during full screen viewing and 37.2mW to 90.7mW of the space on screen is black). The original size test is used to during original size viewing. When compared to the contrast power simulate the small display windows embedded in other consumption shown in Table 2, we propose that: application’s UI’s, e.g., Facebook and YouTube. The power consumption of an AMOLED screen is highly content Table 2 gives the contrast test results, which are measured by dependent and may have relatively little impact on the overall disabling the display output and allowing it to only display a black power consumption of video streaming applications. screen. In these tests, the power consumption of the device is only influenced by the video processing and other background 5.1.2 Video stream decoding cost operations. As expected, the contrast values are not dependent on We also analyzed the power consumption of the video decoding the display content and are almost identical for different content process. The impact of display resolution, frame rate, and encoding samples with the same decoding configuration. By subtracting the bit rate are included in our analysis. power consumption in Table 2 from corresponding items in Table Similar to the experiments presented in Section 4.1.1, two display 1, we can derive the power consumed by the display panel itself. resolutions of 1280x720 (full screen) and 640x360 are adopted in Our results show that the most power consuming video content our tests. For the same display content, bit rate, and fps, the category is Sports. At 60fps and 30fps, the AMOLED screen observed video decoding power consumption is almost identical consumes 29% and 32% of total smartphone power, respectively. when the stream is stretched to full screen from 640x360 and when Such high power consumption comes from the bright color tone and it is viewed at original, letterboxed size. This means that the full complex textures, which require high luminance and balanced color screen pixel stretching in the display buffer consumes very little power for low resolutions. However, a large power rise (~ 100mW) Table 1. Video stream power consumption (mW) is observed if the resolution increases from 640x360 to 1280x720, 1280x720 640x360 640x360 as shown in Table 2. This is due to the significant power required Video (full screen) (full screen) (original size) during the decoding process for high-resolution video. 30fps 60fps 30fps 60fps 30fps 60fps Music 786.4 975.8 682.9 783.7 594.9 698.1 The largest power consumption difference occurs when the frame Sports 960.4 1172.1 855.3 953.2 652.4 751.9 rate changes. The increase in the system power consumption at a Game 869.0 1061.2 786.5 889.1 628.2 730.8 fast frame rate comes from extra workload placed on the CPU. For News 901.3 1124.3 802.7 915.7 612.3 725.5 example, the power consumption of the decoding process at 60 fps Table 2. Contrast power consumption (mW) and a resolution of 640x360 is about 100mW higher than that of one at 30fps. When the resolution rises to 1280x720, the difference 1280x720 640x360 640x360 becomes even more pronounced at 200mW. Video (full screen) (full screen) (original size) 30fps 60fps 30fps 60fps 30fps 60fps We also tested 50 complex CG videos with the same content, time Music 646.9 820.8 543.8 648.2 544.8 649.4 Sports 646.8 823.2 542.5 644.3 548.7 649.0 length, frame rate and resolution but different display quality, i.e., Game 644.3 820.6 543.5 643.7 548.0 648.8 the bit rate. In our tested videos, the bit rate varies from 0.6 Mbps News 647.5 822.1 545.2 644.2 548.2 649.0 to 3.0Mbps. Although the videos share the same resolution, the low-quality ones include many mosaics and the corresponding 3000 video compression can be easier, which led to the lower bit rate. CPU GPU 2500 However, we found that the low bit rate and aggressive Display Base compression do not introduce any significant changes in power 2000 consumption: as is shown in the power track example in Fig. 4, 1500 there are almost no power differences between the different bit rates. Combined with our previous tests, this led us to the 1000 conclusion that higher display resolutions incur a much more 500 significant increase in power consumption than an increased bitrate. Therefore, we propose that: (mW) Consumption Power 0 Galaxy S Nexus S Galaxy S2 Galaxy Galaxy S3 Fast frame rate and high resolution introduce significant video Nexus decoding power consumption. However, when compared to the Fig. 5 Power component breakdown example of video game. whole system, this power portion is not significant. of the game over different hardware models. We chose five different smartphone models to cover the existing GPU/CPU 5.2 Game Power Performance configurations and display panel designs, as shown in Fig. 5. Due to the large volume of graphic computation on CPU/GPU and Although some models share the same chipset, e.g., the popular high display quality requirement, video games have become one of GPUs – PowerVR and Adreno (e.g., Galaxy Nexus uses PowerVR the most power-consuming application types in smartphones. We and ARM9; Galaxy S2 and S3 use Adreno and the CPU from also measured the game power performance, based on the above Qualcomm), a variance in power performance is still observed. For observation, we propose that: example, the power supporting necessary background system Smartphone games consume a large amount of power. This power services varies between 1245mW and 2140mW, which is much is mainly consumed by background computation while significant higher than the normal power level of other applications, such as extra power may be required by user interaction. However, the streaming video players. The ratio between the power consumption AMOLED power consumption is generally small in comparison. of the AMOLED screen (typically 237mW to 363mW) and the whole smartphone reduces to between 15% and 22%. Fig. 5 also We first measured the power consumptions of 20 of the most shows that CPU’s generally consume more power than GPU’s. In popular games from the Play store on the Galaxy S3 the Galaxy S and S2, the contribution of the CPU to total power platform. To enable easy interaction, most of the games are consumption can be up to 40%. composed of big objectives with bright colors. This generates an increase in the power consumption of the AMOLED display. Our 5.3 Camera Power Performance measurements show that the power consumption of the whole Besides streaming video players and video games, camera smartphone is generally within the range of between 1140mW and recording is another important display-related application on 1750mW when user interaction is disabled. smartphones [17]. As has been studied before, camera recording incurs a surprisingly high power cost. However, the performance Some video games require user interaction that involves frequent with the AMOLED display is not yet evaluated in this regard. In sensor operations (e.g., tilt to move or input). For our experiments, we found that: instance, the power consumption of the device while playing Angry Birds and Fruit Ninja was increased to 1640mW and 2220mW, The AMOLED display's power contribution is relatively small respectively, during normal operations. Compared to other compared with that of the rest of the system during camera applications like web browsing (with a power consumption in the recording. Possible power reduction may be achieved by range of 600mW to 1500mW) and video players, these video games optimizing the internal data transformation. require large amounts of power. The display power required with each of another 200 Android video games was simulated with the To arrive at this conclusion, we measured the power consumption Galaxy S3 AMOLED power model. The simulations were made of camera recording applications. Most of the measured with official game trailer videos from YouTube. The average power smartphone cameras have a very high resolution of 6 to 8 consumption of the AMOLED was between 50.3mW~446.4mW. megapixels and are capable of recording HD videos at up to 1080p. Four test modes are included in our measurements: 1) contrast Finally, we analyzed power utilized by background computation mode, where the camera works as normal but the AMOLED screen while playing the open-source video game Quake 3 on the Android is disabled; 2) preview mode, where we only use camera for platform. Although it is an old game, it still offers reasonable game preview but not recording; 3) high quality recording, where the complexity and display quality on present smartphone platform. By video resolution is 1280x720; and 4) low quality recording, where modifying the code, we were able to obtain the power breakdown the video resolution is 640x360. To exclude the power variance

(a) (b) (c) (d) Fig. 6 Galaxy S2 camera recording power histogram: (a) Preview mode; (b) Low quality recording; (c) Zone-in of the power histogram of high quality recording; (d) Zone-in of the power histogram of low quality recording. 3000 Encoding Camera Display Base 2500

2000

1500

1000

500

0

(a) (b) Consumption (mW) Power Galaxy S Nexus S Galaxy S2 Galaxy Galaxy S3 Fig. 7 Galaxy S2 camera recording power histogram: Nexus

(a) 3264x2448; (b) 320x240. Fig. 8 Power component breakdown in camera recording. introduced by the AMOLED screen itself, the smartphone cameras record the same video test benches projected to a screen, including In this paper, we evaluated the power consumption of various pure black, still picture, and the videos from typical categories. display-related applications on smartphones. We first refined the Except for the auto focus, other additional effects are disabled. power analysis of AMOLED display by considering design metrics such as sub-pixel area and matrix. We then conducted the A typical power histogram of Galaxy S2 smartphone in preview AMOLED screen power analysis in the smartphone applications of mode is shown in Fig. 6(a). Note that the power consumption of the a streaming video player, video game, and camera recorder. A wide AMOLED screen can be derived by subtracting the power utilized selection of display content was tested during our experiments during contrast mode from the power consumption in preview running on several representative Samsung smartphone platforms. mode. In recording modes, the power consumption of the We found that the AMOLED screen power model is heavily smartphone increases, as shown in Fig. 6(b). The power variances affected by sub-pixel matrix design while the power efficiency over among the different display contents are barely recognized and the unit area is almost the same. We also found that the power dominated by the video processing with some impacts from consumption of the AMOLED screen while watching a video is recording quality. For a 4-minute camera recording, the average much less than expected while decoding process power power consumption is 1400mW and 1650mW with the low- or consumption is also not significant. In video games, the power high-quality recording, respectively. Out of the total power consumption of the AMOLED screen varies significantly and consumption, the portion of the AMOLED screen accounts for power optimization should focus on the CPU side. Finally, camera 170mW and 180mW in the low- or high-quality recording, recording incurs surprisingly high power consumption, which is respectively. Here the frame rate is 30fps. The power difference constrained by the internal data transformation, with the AMOLED incurred by the resolution up-scaling is 250mW=1650mW– display accounting for only a small portion of overall power cost. 1400mW, which is quite small compared to the overall. 7. REFERENCES However, the zone-ins of the power histograms of each recording [1] J. Huang, et al, “Anatomizing Application Performance Differences mode in Fig. 6(c) and (d) show periodic spike patterns. The highest on Smartphones,” MobiSys, 2010. spikes in the low and high-quality recordings are around 1620mW [2] M. Dong, Y. Kevin, L. Zhong, “Power Modeling of Graphical User and 2050mW, respectively. Also, the spike pattern during low- Interfaces on OLED Displays,” DAC, 2009. quality is more regular than that in high-quality recording. [3] R. Mittal, A. Kansal, R. Chandra, “Empowering Developers to Estimate App Energy Consumption,” MobiCom, 2012. We believe the difference of camera spike patterns in the two types [4] X. Chen, et al, “Quality-retaining OLED Dynamic Voltage Scaling of video recordings is introduced by the constraints on data for Video Streaming Applications on Mobile Device,” DAC, 2012. communication. Since the camera buffer capacity is limited, the [5] M. Dong, L. Zhong, “Chameleon: A Color-Adaptive Web Browser power consumption due to data transfer in the high-quality for Mobile OLED Displays,” MobiSys, 2011. [6] M. Dong, Y. K. Choi, L. Zhong, “Power-Saving Color recording becomes more apparent when the frame size increases. Transformation of Mobile Graphical User Interfaces on OLED-based We also tested the camera recording application with the Displays,” ISLPED, 2009. resolutions of 3264x2448 and 320x240, as shown in Fig. 7. Distinct [7] B. Anand, et al, “Adaptive Display Power management for Mobile spike pattern difference are observed in these two tests. As Games”, MobiSys, 2011. expected, the spike pattern during the 3264x2448 resolution [8] Android PowerManager, http://developer.android.com/. recording becomes prominent. [9] PowerTutor, http://powertutor.org/. [10] C. Yoon, et al, “Appscope: Application Energy Metering Framework The power breakdowns of the camera recording application on for Android Smartphone Using Kernel Activity Monitoring,” USENIX different smartphone models is shown in Fig. 8. The contributors ATC, 2012. to the total power consumption include encoding process, camera [11] L. Zhang, et al, “Accurate Online Power Estimation and Automatic Battery Behavior based Power Model Generation for Smartphones,” device, display and base power consumption. Since the display CODES/ISSS, 2010. content captured from the real environment generally have a very [12] AMOLED, http://en.wikipedia.org/wiki/AMOLED/. low luminance, the power consumption of the AMOLED screen is [13] Mobile Digital Natural Image Engine (MDNIe), only 200mW to 400mW. The power consumption of video http://ww.samsug.com/. encoding, which is greatly impacted by the internal data [14] J. Jacobs, D. Hente, E. Waffenschmidt, “Drivers for ", communication, varies significantly among the different models. Industry Applied Society Annual Meeting, 2007. [15] D. Shin, Y. Kim, N. Chang, M. Pedram, “Dynamic Voltage Scaling Hence, with the tested results, we reached the conclusion that the of OLED Displays,” DAC, 2011. power consumption of the AMOLED display only accounts for a [16] M. Dong, L. Zhong, Z. Deng, “Performance and Power Consumption small part of the device’s overall power consumption when Characterization of 3D Mobile Games,” Computer, 2012. [17] H. Falaki, et al, “Diversity in Smartphone Usage”, MobiSys, 2010. recording video from the camera. [18] OLED Display Technology Shoot-Out, 6. CONCLUSION http://www.displaymate.com/.